GPU share according to Steam Survey (Feb 2022 data)

porina · March 24, 2022

I manually transcribed the currently shown Steam Hardware Survey GPU numbers and used it to obtain the following. Disclaimers: I might have made typos or omissions. Steam Hardware Survey (I'm gonna call it SHS from now) reports values to two decimal places so there will be rounding errors especially on the low end. SHS may be inconsistent in how they present data in different places. For example, there's 0.21% "others" (not Intel, AMD, nvidia) yet Apple have a line at 0.95%. There's an ~11% "others" in the breakdown so that could swing categories depending on what they are. Notably absent are newer AMD GPUs which presumably were individually too small to make the cut, but could still contribute if they would be listed.

Overall reported usage:

nvidia 76%

AMD 15%

Intel 9%

Others within a rounding error - Apple actually get on the list with the G13G which is apparently from the M1.

Anyway, let's break down AMD's GPUs first. We get:

45% Polaris

14% RDNA1

2% RDNA2

15% "Radeon" whatever that is

24% APUs

Given the lack of Vega dGPUs on the list, I have to wonder if that "Radeon" category is due to how the driver reports them and how it is picked up by SHS. Vega APUs are identified and I've put in the APU pile. The RX 590, as the only Polaris card AMD recommends for 1080p output with FSR 2.0, makes up about 5% of Polaris cards overall. RDNA2 is only represented by the 6700 XT. Other models failed to make the cut to be individually listed so this probably under-represents the category.

On Intel, it's all integrated graphics and nothing too interesting there. I do notice the way they're listed, if there might be some double counting of certain models, perhaps if they report back in two ways through the stack. Still, not really significant as everything else will scale accordingly.

On nvidia, my how many models they have!

If we break down by generation we get:

16% Ampere (split 6% mobile, 11% not mobile, yay rounding error)

39% Turing (split 17% RTX, 22% GTX)

32% Pascal

8% Maxwell

5% Older + MX

Despite the shortages there are practically as many Ampere GPUs in use as Turing RTX GPUs. Interesting about a third of Ampere ended up as laptop versions though. Still some decent numbers on Pascal but it looks like Maxwell is falling off now.

As much stick as some give RTX, there are over 50% more of them deployed than all of AMD combined including APUs!